Xuelun Shen

I am a Ph.D. in Computer Science who graduated from Xiamen University. My doctoral supervisor was Cheng Wang.

Email  /  Google Scholar  /  Github

profile photo

Research

I am interested in 3D computer vision, including correspondences between images, camera pose estimation, 3D reconstrucrion, visual localization and SLAM.

GIM: Learning Generalizable Image Matcher From Internet Videos
Xuelun Shen, Zhipeng Cai *, Wei Yin, Matthias Müller, Zijun Li, Kaixuan Wang, Xiaozhi Chen, Cheng Wang*,
† Equal Contribution
ICLR, 2024 (spotlight, top-5%)
project page / online demo / paper / code / youtube / bilibili / blog / zhihu

TL;DR: We propose the first framework to learn a single generalizable image matching model from internet videos.
Application / Influence: DJI Automotive

A Detector-oblivious Multi-arm Network for Keypoint Matching
Xuelun Shen, Qian Hu, Xin Li, Cheng Wang*
IEEE Transactions on Image Processing (TIP), 2023
paper / code

TL;DR: We found that the keypoint score has little effect on the matching performance of SuperGlue, and proposed finding the correspondence between foreground and background in image matching.
Application / Influence: LightGlue

Learning Scale Awareness in Keypoint Extraction and Description
Xuelun Shen, Cheng Wang*, Xin Li, Yifan Peng, Zijian He, Chenglu Wen, Ming Cheng
Pattern Recognition (PR), 2022
paper

RF-Net: An End-to-End Image Matching Network based on Receptive Field
Xuelun Shen, Cheng Wang*, Xin Li, Zenglei Yu, Jonathan Li, Chenglu Wen, Ming Cheng, Zijian He
CVPR, 2019
paper / code

TL;DR: We have applied receptive field constraints in detecting and describing keypoints, and proposed a neighbor ambiguity mask.
Application / Influence: Video Stabilization

H-Net: Neural Network for Cross-domain Image Patch Matching
Weiquan Liu, Xuelun Shen, Cheng Wang*, Zhihong Zhang, Chenglu Wen, Jonathan Li
† Equal Contribution
IJCAI, 2018
paper / code


Real Practical Application

Pure Vision Calibration

Multi-Pass Reconstruction / Image Matching in Challenging Conditions


Design and source code from Jon Barron's website